RynnUnify 16B model unifies embodied understanding, planning, and control
Original title: RynnUnify: Towards Unified Embodied Brain
Authors report a 9.75-point Rynn-VP gain over Cosmos 3 Nano and 46.7% average success on three real-robot tasks.
The paper introduces RynnUnify, a 16B unified embodied model that interleaves video and language for spatial reasoning, physical prediction, grounding, and long-horizon visual planning. An agentic reinforcement-learning extension selectively uses video predictions for unobserved states, and a world-action extension maps the model to robot control. New benchmarks Rynn-VP and Rynn-GEU evaluate planning and anticipatory reasoning. Reported results include the highest average embodied-understanding score among compared unified models.
Unconfirmed: Scores and success rates are the authors' reported experimental results.
- RynnUnify has 16B parameters and uses interleaved video-language reasoning.
- The agentic extension improves the Rynn-GEU average by 6.0 points over direct reasoning.
- World-action success is 51.5% on 20 selected composite RoboCasa365 tasks.
- Average success is 46.7% across three real-robot tasks.
- AI models
- RynnUnifyCosmos 3 Nano
From the original article
1 min read at huggingface.co
We introduce RynnUnify, a 16B unified embodied model that combines strong embodied understanding with long-horizon visual planning through interleaved video--language reasoning. With capabilities spanning spatial reasoning, physical prediction, and grounding, RynnUnify connects an understanding of the physical world with visual anticipation of how it evolves across successive subtasks.
By Minghao Zhu, Bohan Hou, Kun Xiang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.